Investing Beyond HBM: High-Bandwidth Flash, Sandisk, and the Next Memory Cycle in AI

Author: Zion Zhao Real Estate | 88844623 | 狮家社小赵 | wa.me/6588844623

Author’s note and disclaimer: For general education and market literacy only. Not financial, investment, legal, accounting, or tax advice, and not an offer, solicitation, or recommendation. Information is general and may be inaccurate or change. No liability accepted. Investing involves risk, including loss of principal; past performance is not indicative of future results. 

The Next Memory Cycle: High-Bandwidth Flash, Sandisk, and the New Economics of AI Inference

The investment debate around High-Bandwidth Flash, or HBF, is not really about whether it will replace High-Bandwidth Memory, or HBM. That is the wrong question. The more important question is whether AI inference is forcing the industry to create an entirely new memory tier between premium DRAM and conventional storage. On that front, HBF deserves serious attention. The case for it is not speculative fiction. It is rooted in a very real architectural problem: the memory wall. In modern AI systems, especially large language model inference, the bottleneck is increasingly memory capacity, memory bandwidth, and data movement rather than raw compute alone (Gholami et al., 2024; Ma & Patterson, 2026). (arXiv)

That is why HBF matters. Sandisk and SK hynix are explicitly positioning it not as a direct HBM substitute, but as a new layer between HBM and SSDs for the AI inference era. In practical terms, this means trying to deliver far more memory capacity than HBM can economically provide, while still offering materially higher effective performance than traditional NAND storage. Sandisk’s own published HBF materials describe a product concept aimed at up to 8 to 16 times the capacity of HBM, with first-generation targets of 1.6 TB/s read bandwidth and 512 GB per 16-die stack. The company has also argued, based on internal simulation, that HBF can approach the performance of unlimited-capacity HBM for certain inference tasks while doing so at similar cost. That is precisely why the market is paying attention. HBF is not about dethroning HBM. It is about making the overall memory hierarchy more economically scalable for inference (Sandisk, 2025a, 2025b; SK hynix, 2026). (SanDisk Documents)

The technology stack behind this idea is also more substantial than the headlines suggest. HBF is being built on top of the Sandisk and Kioxia flash ecosystem, particularly BiCS FLASH and CMOS directly Bonded to Array, or CBA, technologies. Kioxia and Sandisk’s February 2025 announcement on next-generation 3D flash highlighted 4.8 Gb/s NAND interface speed, improved power efficiency, and higher density, while Kioxia’s own technical materials explain that CBA allows separate optimization of the CMOS and memory-cell wafers before bonding, improving performance, power, area, and cost characteristics. In other words, HBF is not being invented from scratch. It is an architectural extension of decades of 3D NAND and packaging progress, repurposed for a new AI use case (Kioxia, 2025a, 2025b; Sandisk, 2025c). (KIOXIA Corporation)

Still, investors should keep their enthusiasm disciplined. Technical plausibility is not the same as near-term earnings power. Sandisk’s public roadmap has pointed to customer sampling beginning in 2026, while the February 2026 standardization push with SK hynix shows that the ecosystem is still in the process of defining the framework under the Open Compute Project. That is encouraging, but it is not the same as large-scale deployment or material profit contribution. Even SK hynix’s own messaging implies that demand for more complex solutions like HBF is part of a longer-duration shift tied to the rise of inference infrastructure. Put simply, this story is real, but it is early (Sandisk, 2025b; Sandisk & SK hynix, 2026; SK hynix, 2026). (Sandisk)

That distinction matters for the investment thesis. In the near term, Sandisk remains a flash and NAND-cycle story first, with HBF functioning more as strategic optionality than as the immediate driver of results. Sandisk’s fiscal fourth quarter 2025 results still reflected the conventional realities of this business, including bit shipments, pricing, margins, technology ramp, and market mix. The company reported $1.90 billion in quarterly revenue, up 12 percent sequentially, with BiCS8 ramp and customer qualifications underway. So the right way to think about Sandisk today is not that HBF has already arrived in the income statement. It is that Sandisk now offers one of the clearest public market options on a possible new memory tier for AI inference, on top of its existing participation in the NAND cycle (Sandisk, 2025d). (Sandisk Corporation)

The broader takeaway is this: the next memory cycle may not be defined by a winner-take-all contest between HBM and flash. It may instead be defined by a smarter hierarchy, where premium bandwidth, scalable capacity, and system economics are rebalanced for inference. If that happens, HBF will matter not because it kills HBM, but because it makes AI infrastructure more deployable at scale. For investors, that is the real signal. The future may belong not to a single memory technology, but to the companies that can best architect the layers around the memory wall (Gholami et al., 2024; Ma & Patterson, 2026; Sandisk & SK hynix, 2026). (arXiv)

References
Gholami, A., Yao, Z., Kim, S., Hooper, C., Mahoney, M. W., & Keutzer, K. (2024). AI and memory wall. IEEE Micro.

Kioxia. (2025a, February 20). Kioxia and Sandisk unveil next-generation 3D flash memory technology achieving 4.8Gb/s NAND interface speed.

Kioxia. (2025b). What is CMOS directly Bonded to Array (CBA) technology?

Ma, X., & Patterson, D. (2026). Challenges and research directions for large language model inference hardware(arXiv:2601.05047).

Sandisk. (2025a, July). High-Bandwidth Flash fact sheet.

Sandisk. (2025b). Scaling beyond the wall: Inside Sandisk’s High Bandwidth Flash for AI.

Sandisk. (2025c, February 19). Kioxia and Sandisk unveil next-generation 3D flash memory technology.

Sandisk. (2025d, August 14). Sandisk reports fiscal fourth quarter 2025 financial results.

Sandisk. (2026, February 25). Sandisk and SK hynix begin global standardization of next-generation memory solution, High Bandwidth Flash (HBF).

SK hynix. (2026, February 25). SK hynix and Sandisk begin global standardization of next-generation memory “HBF”.

High-Bandwidth Flash and the Future of AI Infrastructure: Why Sandisk Could Matter More Than Investors Think

High-Bandwidth Flash is not a direct High-Bandwidth Memory replacement, but a promising new inference-era memory tier. If commercialised successfully, it could ease the artificial intelligence memory wall, improve capacity economics, and reshape the next memory cycle, benefiting Sandisk, its ecosystem partners, and semiconductor equipment leaders alike.

In today’s world, real estate should not be viewed in isolation. The strongest property decisions are increasingly made at the intersection of macroeconomics, geopolitics, capital flows, interest rates, equity markets, currency trends, regulatory policy, and long-term wealth preservation. That is precisely why choosing the right real estate advisor matters.

As a Singapore real estate agent, I believe my role goes far beyond arranging viewings or negotiating transactions. I dedicate hours every single day to studying the market, writing in-depth essays, analysing macroeconomic developments, tracking global capital trends, and conducting proper due diligence across property, equities, cryptocurrency, and broader asset allocation themes. My approach is built on discipline, research, and constant learning. It is shaped not only by real estate experience, but also by my understanding of economics, global affairs, portfolio construction, Singapore land and business law, and a leadership background that values preparation, responsibility, and clarity under pressure.

For international buyers, China Chinese families, Southeast Asian investors, Singapore-based clients, ultra high net worth individuals, family offices, institutional investors, and families planning immigration or overseas education pathways, this matters. You should not rely on an agent who only understands square footage and brochures. You need someone who can help you see how Singapore property fits into a broader portfolio strategy, how to position capital across cycles, how to think about risk, and how to identify opportunities in a stable, globally respected market.

Singapore real estate can play an important role in a well-constructed portfolio. Compared with more volatile asset classes, quality property can offer relative stability, long-term capital appreciation, and rental income that functions much like a dividend-producing asset. For many investors, it is not just a home or a purchase. It is a strategic store of value, an income-generating asset, and a cornerstone of intergenerational wealth planning.

If you value an advisor who stays closely abreast of themes such as the next memory cycle, artificial intelligence infrastructure, international geopolitics, macroeconomics, capital markets, and cross-asset investing, while also understanding the realities of Singapore property on the ground, I would be honoured to assist you.

Whether you are looking to buy, sell, rent, invest, relocate, preserve wealth, or position your family for the future through Singapore, let us have a thoughtful conversation. In a complex world, you deserve a real estate advisor who does the homework, sees the bigger picture, and helps you make property decisions with confidence, prudence, and conviction.